Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TangledFeatures: Robust Feature Selection in Highly Correlated Spaces

Allen Daniel Sunny

TL;DR

The paper targets feature selection in highly correlated predictor spaces typical of structural biology, introducing TangledFeatures, a stability-based, pre-hoc pipeline that clusters correlated features, selects robust cluster representatives, and refines them to preserve predictive signal. It maps intra-atomic distances to backbone torsions $\phi$ and $\psi$ via cosine–sin representations, enabling stable interpretation of structural drivers. Through experiments on Alanine dipeptide, TangledFeatures demonstrates strong predictive performance with markedly improved stability and interpretability, aligning selected features with meaningful intra-atomic distances. The work provides an open-source R package to facilitate reproducible, redundancy-aware feature selection in structured biological data.

Abstract

Feature selection is a fundamental step in model development, shaping both predictive performance and interpretability. Yet, most widely used methods focus on predictive accuracy, and their performance degrades in the presence of correlated predictors. To address this gap, we introduce TangledFeatures, a framework for feature selection in correlated feature spaces. It identifies representative features from groups of entangled predictors, reducing redundancy while retaining explanatory power. The resulting feature subset can be directly applied in downstream models, offering a more interpretable and stable basis for analysis compared to traditional selection techniques. We demonstrate the effectiveness of TangledFeatures on Alanine Dipeptide, applying it to the prediction of backbone torsional angles and show that the selected features correspond to structurally meaningful intra-atomic distances that explain variation in these angles.

TangledFeatures: Robust Feature Selection in Highly Correlated Spaces

TL;DR

The paper targets feature selection in highly correlated predictor spaces typical of structural biology, introducing TangledFeatures, a stability-based, pre-hoc pipeline that clusters correlated features, selects robust cluster representatives, and refines them to preserve predictive signal. It maps intra-atomic distances to backbone torsions and via cosine–sin representations, enabling stable interpretation of structural drivers. Through experiments on Alanine dipeptide, TangledFeatures demonstrates strong predictive performance with markedly improved stability and interpretability, aligning selected features with meaningful intra-atomic distances. The work provides an open-source R package to facilitate reproducible, redundancy-aware feature selection in structured biological data.

Abstract

Feature selection is a fundamental step in model development, shaping both predictive performance and interpretability. Yet, most widely used methods focus on predictive accuracy, and their performance degrades in the presence of correlated predictors. To address this gap, we introduce TangledFeatures, a framework for feature selection in correlated feature spaces. It identifies representative features from groups of entangled predictors, reducing redundancy while retaining explanatory power. The resulting feature subset can be directly applied in downstream models, offering a more interpretable and stable basis for analysis compared to traditional selection techniques. We demonstrate the effectiveness of TangledFeatures on Alanine Dipeptide, applying it to the prediction of backbone torsional angles and show that the selected features correspond to structurally meaningful intra-atomic distances that explain variation in these angles.
Paper Structure (14 sections, 3 equations, 5 figures, 3 tables)

This paper contains 14 sections, 3 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Overview
  4. Clustering
  5. Selection Module
  6. Refinement Module
  7. Objective
  8. Experiments
  9. Predictive Accuracy
  10. Stability
  11. Interpretability
  12. Conclusion
  13. Software Availability
  14. Selected Distances

Figures (5)

  • Figure 1: Illustration of the Ramachandran plot, showing allowed conformational regions for the backbone torsional angles $\phi$ and $\psi$. In this work, $\phi$ (C–N–C$_\alpha$–C) and $\psi$ (N–C$_\alpha$–C–N) serve as prediction targets, while intra-atomic distances are used as input features.
  • Figure 2: Stability for the $\phi$ torsional angle. (a) Kuncheva index, (b) Spearman rank correlation.
  • Figure 3: Comparison of selected feature subsets for the $\phi$ torsional angle.
  • Figure : Feature selection stability for $\phi$ torsional angle
  • Figure : Feature selection stability for $\psi$ torsional angle